lapis-hong / wide_deep Goto Github PK

I see you process multivalue features in dataset.py, but not used in feature column, is that right?
How can the multivalue features be used ?

waiting for the response~
Thanks a lot!

The data input need a ordered schema. While you use yaml.load() to load the schema, which return a dict. How can you keep the right order of the feature names to read the corret columns in the input data?

Hello,I found a performance issue in the definition of input_fn ,
Lapis-Hong/wide_deep/blob/master/python/lib/utils/create_record.py,
dataset = dataset.map(_parse_function) was called without num_parallel_calls.
I think it will increase the efficiency of your program if you add this.

Here is the documemtation of tensorflow to support this thing.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

when i use multivalue feature, such as [['a', 'b', 'c'], ['a', 'b']].
the program will report bug like: cannot batch elements of different shape. element 0 has shape [3], but element 1 has shape [2].
i know the different shape of multivalue feature result in this bug.
i think the function Dataset.padded_batch maybe solve the problem, i write like dataset = dataset.padded_batch(5, padded_shapes=[None]), it reported bug TypeError: If shallow structure is a sequence, input must also be a sequence. Input has type: list
so, how can i process the multivalue exactly?
i am eager to solve this problem, can anyboby help me? thanks a lot!

hi, in the tf servering client.py , when the input data one feature is need split (like the input_fn function in dataset.py) , like 'a#b#c' , how can we deal with it?
thanks for answering.

The current client.cc seems like a demo for image classificaiton? can you provide a corresponding c++ demo for the wide_deep model？
I didn't know how to prepare the input for this case.
** How to process the raw feature same as the feature_column（training process） in c++ prediction? **